[Prev][Next][Index]

CSLI Calendar, 30 March 1995, vol.10:21




        C S L I   C A L E N D A R   O F   P U B L I C   E V E N T S
______________________________________________________________________________

30 March 1995                    Stanford                      Vol. 10, No. 21
______________________________________________________________________________

      A weekly publication of the Center for the Study of Language and
Information (CSLI), Stanford University, Ventura Hall, Stanford, CA 94305-4115
                               ____________

             CSLI ACTIVITIES DURING 29 MARCH -- 7 APRIL 1995

  WEDNESDAY, 29 MARCH
         4:00 - Seminar on Computational Learning and Adaptation
                Cordura Hall, Room 100
                Nonlinear Mixture Models for Time Series Analysis:
                Discovering Regimes and Avoiding Overfitting
                Andreas S. Weigund, Colorado Computer Science
                Abstract below

  MONDAY, 3 APRIL
         3:15 - Special Seminar
                Ventura Hall, Room 17
                The Euclidean Diagram
                Kenneth Manders, Pittsburgh Philosophy
                Abstract below

  THURSDAY, 6 APRIL
        12:15 - CSLI Seminar
                Series on Intelligent Agents
                Cordura Hall, Room 100
                Small Mobile Robots: Experiments with ERRATIC
                Kurt Konolige, SRI International
                Abstract below

         2:15 - CSLI Seminar
                Series on Visual Programming
                Cordura Hall, Room 100
                IC-By-Example: Delivering Programming Capabilities
                to the Uninitiated
                Moshe Zloof, Hewlett-Packard
                Abstract below

         4:15 - SSP Forum
                Building 60, Room 61-F
                Making a Mind vs. Modeling the Brain: AI Back at a Branchpoint
                Hubert Dreyfus, UC Berkeley Philosophy
                Abstract below

  FRIDAY, 7 APRIL
         3:30 - Linguistics Colloquium
                Cordura Hall, Room 100
                Title to be announced
                Martin Kay, Xerox PARC and Stanford Linguistics

        7:00pm- Linguistics Conference
                Building 420, Lower Level Auditorium
                27th Annual Child Language Research Forum
                Partial schedule below

                               ____________

The CSLI Calendar appears on Thursday of each week throughout the academic
year.  Announcements, abstracts, and other information to appear in the
Calendar on a given Thursday should be submitted by e-mail to
<incalendar@csli.stanford.edu> by 5:00 p.m. on the previous Tuesday.

Past issues of the CSLI Calendar, a quarterly schedule of upcoming CSLI
events, and other information about CSLI are available on the Web, URL
<http://www-csli.stanford.edu/>.  The Calendar, with available abstracts, is
also posted each week to the csli.bboard newsgroup.

                               ____________

             SEMINAR ON COMPUTATIONAL LEARNING AND ADAPTATION
                          on Wednesday, 29 March
                    4:00 p.m., Cordura Hall, Room 100
            Nonlinear Mixture Models for Time Series Analysis:
               Discovering Regimes and Avoiding Overfitting
                            Andreas S. Weigund
                        Colorado Computer Science
                        <andreas@cs.colorado.edu>

When trying to forecast the future behavior of real-world systems, two of the
key problems are overfitting (particularly serious for noisy processes) and
regime switching (the underlying process changes its characteristics).  In
this talk we show how Gaussian mixture models with nonlinear experts point to
solutions to these problems.  In connectionist terms, the architecture
consists of several experts and a gating net.  The nonlinear experts put out
the conditional mean (as usual), but each expert also has its own adaptive
width.  The gating net puts out an input-dependent probability for each expert.
There is a supervised component in learning, to predict the next value(s), and
an unsupervised component, to discover the (hidden) regimes.  We report a
number of results:

(1) the gating net discovers the different regimes that underlie the 
    process: its outputs segment the data correctly into the different 
    regions.

(2) the widths associated with each expert characterize the sub-processes:
    i.e., the variances give the expected squared error for each regime.

(3) there is significantly less overfitting compared to single nets, 
    for two reasons: only subsets of the potential inputs are given to 
    the experts and gating net, and the experts learn to match their 
    variances to the (local) noise levels, thus only learning as much 
    as the data support.

We compare these results from the mixture model to single networks of
different sizes, as well as to nets with two outputs, one for the mean, the
other one for the confidence interval as a function of the input.  Several
data sets are used: a computer-generated series, the laser data set from the
Santa Fe Competition, the daily electricity demand of France, and the daily
exchange rate between German marks and US dollars.

The goal of this seminar is to increase communication among local researchers
with interests in computational approaches to learning and adaptation. If you
would like to be added to (or removed from) the mailing list, or if you are
interested in giving a talk in the seminar, please send email to
<langley@cs.stanford.edu>.

                               ____________

                             SPECIAL SEMINAR
                            on Monday, 3 April
                    3:15 p.m., Cordura Hall, Room 100
                          The Euclidean Diagram
                             Kenneth Manders
                          Pittsburgh Philosophy
                       <mandersk@vms.cis.pitt.edu>

In traditional "Euclidean" geometry, relationships in diagrams may license
inference, just as relationships in the text.  According to the view now
established, this is a defect of rigor.

But Euclidean practice had an extraordinary career.  It was a stable and
fruitful tool of investigation across diverse cultural contexts for over two
thousand years.  It generally struck knowledgeable people as the most rigorous
of human ways of knowing, even in the face of centuries of internal criticism
in antiquity.

We reconstruct the role of diagrams in Euclidean inference, seeking a more
accurate view of the strengths and weaknesses of traditional geometrical
argument.

Traditional geometrical argument co-ordinates two means of expression with
very different characteristics, diagrams and text.  Understanding their
relative strengths brings out forces for change in intellectual practices.
These phenomena bring out possibilities for representational contrasts which
philosophers seem to have ignored.

A renewed and tightened philosophical grip on traditional geometry may also
help interpret thought shaped in part by reflection on that practice
(Descartes, Kant?)

                               ____________

                               CSLI SEMINAR
                       SERIES ON INTELLIGENT AGENTS
                           on Thursday, 6 April
                    12:15 p.m., Cordura Hall, Room 100
              Small Mobile Robots: Experiments with ERRATIC
                              Kurt Konolige
                            SRI International
                          <konolige@ai.sri.com>

For the past two years I have taught a course at Stanford in building and
using small mobile robot platforms, aptly named ERRATIC.  The robots are
home-grown, built using sheet metal, model airplane parts, Polaroid camera
sonars sensors, and a small microprocessor.  They are capable of supporting
experiments in sensor interpretation, motion control, planning and
exploration.  In the course of writing robot control programs, we worked on
the basic problems of agents acting in real-world environments: how to move
without bumping into things, how to achieve goals, and how to map the
environment and keep track of position within the map.  This talk focuses on
some of the experiments that succeeded, and more importantly, those that were
only partly successful.  I will bring one of the robots to the talk.

NOTE: A complete schedule for the CSLI seminar series on intelligent agents is
available, via your favorite web browser, at <http://www-csli.stanford.edu/>,
or <http://www-csli.stanford.edu/csli/intellagents.html>, to be more specific.
The aim of this series of talks is to examine some different conceptions of
"intelligent agents" as developed and exploited in robotics and software
design.  A broad range of perspectives will be presented throughout the
quarter -- from open agent architectures, adaptive Web browsers,
resource-constrained rational agents, and net-based information brokers, to
intelligent pilot agents for virtual combat environments, teleo-reactive
planners and learners, behavior-based robotics, behavior-based robotics, and a
real though perhaps erratic physical robot or two.

These talks are scheduled for 12:15 (noonish) on Thursdays throughout the
Spring quarter at Stanford, starting April 6, in Cordura Hall, Room 100.  CSLI
will provide a sandwich and beverage (for $4.50, paid at the door) if you
order beforehand.  Send sandwich orders (turkey, ham, roastbeef, or
vegetarian) to <sanchez@csli.stanford.edu> some time on or before Tuesday of
the week of the given talk.

                               ____________

                               CSLI SEMINAR
                       SERIES ON VISUAL PROGRAMMING
                           on Thursday, 6 April
                    2:15 p.m., Cordura Hall, Room 100
                  IC-By-Example: Delivering Programming
                      Capabilities to the Uninitiated
                              Moshe M. Zloof
                       Hewlett-Packard Laboratories
                            <zloof@hpl.hp.com>

In an effort to make computers more pervasive, we propose to empower ordinary
people with the ability to program their own applications.  Through a
graphical language called _Interoperation-&-Customization-By-Example_ (ICBE),
the user can express a "program" in a "what you want" fashion while leveraging
on existing applications.  Just as Relational Technology brought the "what you
want" paradigm to database query, we are trying to deliver similar promise in
the larger context of building applications.

Subtopics of the talk:
  *  High-level declarative based on domain calculus (By-Example).
  *  Programming is accomplished by stating "what you want" in a
     picture-like fashion. 
  *  Leveraging on existing application (in particular Databases).
  *  End-user customization by interactive programming.
  *  Enables small clientele applications otherwise not economically
     feasible at today's programming costs.

                               ____________

                          SYMBOLIC SYSTEMS FORUM
                           on Thursday, 6 April
                    4:15 p.m., Building 60, Room 61-F
                  Making a Mind vs. Modeling the Brain:
                         AI Back at a Branchpoint
                              Hubert Dreyfus
                          UC Berkeley Philosophy
 
In the nineteen fifties two opposed research programs emerged, each seeking to
use computers to produce intelligence.  One saw computers as a system for
manipulating mental symbols; the other, as a medium for modeling the brain.
One sought to use computers to instantiate a formal representation of the
world; the other, to simulate the interactions of neurons.  One took problem
solving as its paradigm of intelligence; the other, learning.  One utilized
logic, the other statistics.  One school was the heir to the rationalist,
reductionist tradition in philosophy; the other viewed itself as idealized,
holistic neuro-science.  By the late sixties, however, the neural net approach
had been almost completely eliminated and the symbolic approach remained,
until recently, the only game in town.

My talk examines the philosophical assumptions underlying the atomistic,
symbolic research program which led to the temporary defeat of the holistic,
neural network approach.  It then reviews the unexpected and unresolved
problems facing the symbolic approach, and suggests that the underlying
assumption that there must be a theory of every domain has been disconfirmed.
It is thus time for neural network modeling, which is not committed to the
rationalistic assumptions of the symbolic approach, to have its day.  However,
given the likely importance of the particular architecture of the human neural
network, as well as the role of emotions, culture, and bodies in the way human
beings come to embody an understanding of the world, neural network modeling,
while correct in principle, may well fail as a program for giving computers
general intelligence.

                               ____________

                          LINGUISTICS COLLOQUIUM
			    on Friday, 7 April
                    3:30 p.m., Cordura Hall, Room 100
                          Title to be announced
                                Martin Kay
                   Xerox PARC and Stanford Linguistics
                           <kay.parc@xerox.com>

                               ____________

                          LINGUISTICS CONFERENCE
                   Building 420, Lower Level Auditorium
                27th Annual Child Language Research Forum
                            April 7-8-9, 1995
             <http://www-csli.stanford.edu/csli/clrf27.html>

FRIDAY, 7 APRIL

7.00-7.30 pm    Registration.
7.30 pm         Introductory Remarks: Eve V. Clark.
7.40 pm         Panel: "Current approaches to phonological development"
                Organizer: Katherine Demuth.
9.45 pm         Reception, Rm 146 Margaret Jacks Hall.

SATURDAY, 8 APRIL

8.30 am         Registration.
9.00-10.30 am   Paper Session: Shanley E. M. Allen; Clifton Pye, Diane Loeb,
                and Yin-Yin Pao; Charlotte Koster and Sjoukje van der Wal.
10.30-11.00 am  Break
11.00-12.30 pm  Paper Session: Elena V. M. Lieven; Evelien Krikhaar and Frank
                Wijnen; Katharina Boser.
12.30-2.00 pm   Lunch
1.30-3.00 pm    Poster Session
2.30-3.30 pm    Paper Session: Sharon Armon-Lotem; Xiangdong Jia, Patricia J.
                Brooks, and Martin D. S. Braine. 
3.30-4.00 pm    Break
4.00-6.00 pm    Workshop: "Evaluative Elements in Children's Narratives"
                Organizers: Ruth Berman and Judith Reilly.

SUNDAY, 9 APRIL

9.00-10.30 am   Paper Session: Hilke Elsen; Catalina M. Johnson; Melissa
                Bowerman, Lourdes de Leon, & Soonja Choi
10.30-11.00 am  Break
11.00-12.00 pm  Paper Session: Marie E. Helt and Susan H. Foster-Cohen; Elaine
                S. Andersen, Beatrice DuPuy, and Maquela Brizuela. 

All the meeting sessions will be held in the Lower Level Auditorium, Jordan
Hall (Psychology Department), Stanford University.  This building is located
at the head of the Oval, to the right along the front of the Quad.

                               ____________

                           COURSE ANNOUNCEMENT
                           Spring Quarter 1995
                  Philosophy 382, Philosophy of Language
                        Thursdays, 3:15--5:05 p.m.
                          Cordura Hall, Room 104
                       Instructor: Julius Moravcsik
                        <julius@csli.stanford.edu>

This seminar will be concerned with Chomsky's latest proposals concerning:
 (a) Why we should not include matters of reference and denotation in everyday
     use as part of linguistic competence (and hence in a semantic theory of
     natural languages in non-scientific use).
 (b) Chomsky's sketches of arguments that would show the limitations of humans
     understanding their own faculty of knowledge and understanding.

We will see that these claims open up problems concerning:
  i) Syntax in logic versus syntax in natural languages.
 ii) The relationship between everyday use of natural languages and "the
     language of science."
iii) Arguments purporting to show that a natural language like English cannot
     be a formal language in Tarski's sense.
 iv) Proposals for a "generative conception" of the lexicon.

We will consider mimeographed material from Chomsky, Putnam, Pustejovsky, and
Moravcsik.

I have to know roughly how many will attend fairly regularly so that (a) I can
be sure the room is adequate and (b) I can guess how many copies of the
unpublished materials need to be produced.  So please reply to this
announcement (<julius@csli>) if you intend to come fairly regularly.

We will start by surveying the conceptions of syntax in twentieth century
philosophy of language, then move on to consider arguments showing that, e.g.,
English is not a "formal language."  We will then see how these arguments
affect the scientific/non-scientific distinction, and how the presence of
functionality in natural language is related to the purported unknowability
arguments.  If a volunteer knows the Penrose argument, we will consider that
also, provided that the volunteer can make it intelligible to all of us.

                               ____________